Elevated CO2 maintains grassland net carbon uptakeunder a future heat and drought extremeJacques Roya,1,2, Catherine Picon-Cochardb,1, Angela Augustib,c, Marie-Lise Benotb,d, Lionel Thieryb, Olivier Darsonvilleb,Damien Landaisa, Clément Piela, Marc Defosseza, Sébastien Devidala, Christophe Escapea, Olivier Ravela,Nathalie Fromine, Florence Volairee,f, Alexandru Milcua,e, Michael Bahng, and Jean-François Soussanab

aEcotron Européen de Montpellier, Unité Propre de Service 3248, Centre National de la Recherche Scientifique (CNRS), Campus Baillarguet, F-34980Montferrier-sur-Lez, France; bGrassland Ecosystem Research, Unité de Recherche 874, Institut National de la Recherche Agronomique (INRA), F-63039Clermont-Ferrand, France; cInstitute of Agroenvironmental and Forest Biology, Consiglio Nazionale delle Ricerche, 2-05010 Porano (TR), Italy; dBiodiversitéGènes et Communautés, INRA, Université de Bordeaux, F-33615 Pessac, France; eCentre d’Ecologie Fonctionnelle et Evolutive, CNRS, Unité Mixte deRecherche 5175, Université de Montpellier, Université Paul Valéry, École Pratique des Hautes Études, F-34293 Montpellier Cedex 5, France; fUnité SousContrat 1338, INRA, Centre d’Ecologie Fonctionnelle et Evolutive F-34293 Montpellier Cedex 5, France; and gInstitute of Ecology, University of Innsbruck,A-6020 Innsbruck, Austria

Edited by William H. Schlesinger, Cary Institute of Ecosystem Studies, Millbrook, NY, and approved April 11, 2016 (received for review December 12, 2015)

Extreme climatic events (ECEs) such as droughts and heat wavesare predicted to increase in intensity and frequency and impact theterrestrial carbon balance. However, we lack direct experimentalevidence of how the net carbon uptake of ecosystems is affected byECEs under future elevated atmospheric CO2 concentrations (eCO2).Taking advantage of an advanced controlled environment facilityfor ecosystem research (Ecotron), we simulated eCO2 and extremecooccurring heat and drought events as projected for the 2050sand analyzed their effects on the ecosystem-level carbon and waterfluxes in a C3 grassland. Our results indicate that eCO2 not onlyslows down the decline of ecosystem carbon uptake during theECE but also enhances its recovery after the ECE, as mediated byincreases of root growth and plant nitrogen uptake induced by theECE. These findings indicate that, in the predicted near future climate,eCO2 could mitigate the effects of extreme droughts and heat waveson ecosystem net carbon uptake.

climate change | extreme events | elevated CO2 | carbon fluxes |grassland ecosystem

Increased aridity and heat waves are projected to increase in the21st century for most of Africa, southern and central Europe,

the Middle East, and parts of the Americas, Australia, and south-east Asia (1–3). These regions have a large fraction of their landcovered by grasslands and rangelands, a biome covering approxi-mately one-quarter of the Earth’s land area and contributing to thelivelihoods of more than 800 million people (4). There is mountingevidence that extreme climatic events (ECEs) may significantly af-fect the regional and global carbon (C) fluxes (3, 5–9) and poten-tially feed back on atmospheric CO2 concentrations and the climatesystem (7). However, our knowledge concerning the outcome of theinteraction between future ECEs and elevated atmospheric CO2concentrations (eCO2) for ecosystem C stocks is equivocal (10–12).Studies focusing on plant physiological responses have shown thateCO2 has the potential to mitigate future drought-related stress onplant growth by reducing stomatal conductance, thereby increasingwater use efficiency (WUE) (13–15) and preserving soil moisture(16–18). However, to date, little is known on whether and howeCO2 alters the consequences of ECEs for ecosystem net C uptake.Because the capacity of ecosystems to act as a C sink depends onthe relative effects of eCO2, ECE, and their potential interaction onboth plant and soil processes, an integrated assessment of all Cfluxes during and after the ECEs is important if we are to estimatethe overall C balance.Using the Montpellier CNRS Ecotron facility (www.ecotron.

cnrs.fr), we tested with 12 large controlled environment units(macrocosms, SI Appendix, Fig. S1) whether (i) an ECE (severedrought and heat wave) predicted for the 2050s reduces ecosystemnet C uptake by reducing ecosystem photosynthesis relative toecosystem respiration (Reco), (ii) eCO2 buffers the negative effects

of the ECE on ecosystem CO2 fluxes and increases ecosystemwater-use efficiency during the ECE, and (iii) eCO2 speeds upthe recovery of ecosystem C uptake after the ECE. We exposed aseminatural upland grassland (botanical composition see SIAppendix, Table S1) from the French Massif Central (45°43′N,03°01′E, 800 m above sea level) to the average climatic condi-tions for the 2050s as projected by the downscaled ARPEGEv4climate model (19). We exposed large (4 m2 and 0.6 m depth)and intact ecosystem monoliths in lysimeters to (i) ambient CO2(aCO2 at 390 μmol·mol−1) and predicted elevated levels of at-mospheric CO2 concentrations (eCO2 at 520 μmol·mol−1) and(ii) the presence/absence of the most severe drought and heatwave (ECE) projected by the above-mentioned climate modelover the 2050s. The experiment was based on 12 macrocosmexperimental units, using a two factorial crossed design (elevatedCO2 and ECE, each with two levels) with three replicates pertreatment combination. The ECE included a reduction in pre-cipitation by 50% during 4 wk in midsummer, followed by 15 d withno precipitation and a concomitant increased temperature by+3.4 °C(Fig. 1). Thereafter, the precipitation was gradually increasedduring 26 d (see Methods for details). In addition to vegetationcharacteristics, we monitored the net ecosystem CO2 exchange

Significance

Ecosystems are responding to climate change and increasingatmospheric CO2 concentrations. Interactions between thesefactors have rarely been assessed experimentally during andafter extreme climate events despite their predicted increase inintensity and frequency and their negative impact on primaryproductivity and soil carbon stocks. Here, we document how agrassland exposed to a forecasted 2050s climate shows a re-markable recovery of ecosystem carbon uptake after a severedrought and heat wave, this recovery being amplified underelevated CO2. Over the growing season, elevated CO2 entirelycompensated for the negative impact of extreme heat anddrought on net carbon uptake. This study highlights the im-portance of incorporating all interacting factors in the predic-tions of climate change impacts.

Author contributions: J.-F.S. designed research; J.R., C.P.-C., A.A., M.-L.B., L.T., O.D., D.L., C.P.,S.D., C.E., O.R., N.F., and F.V. performed research; J.R., C.P.-C., A.A., D.L., C.P., M.D., N.F., F.V.,and A.M. analyzed data; and J.R., C.P.-C., A.M., M.B., and J.-F.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1J.R. and C.P.-C. contributed equally to this work.2To whom correspondence should be addressed. Email: jacques.roy@ecotron.cnrs.fr.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1524527113/-/DCSupplemental.

6224–6229 | PNAS | May 31, 2016 | vol. 113 | no. 22 www.pnas.org/cgi/doi/10.1073/pnas.1524527113

Dow

nloa

ded

by g

uest

on

May

25,

202

0

(NEE) and ecosystem evapotranspiration (ET). Furthermore,we derived values of gross primary productivity (GPP), Reco, andWUE for the prestress, stress, and poststress periods.

ResultsDuring the prestress period in spring, ecosystem CO2 and watervapor fluxes increased with the development of the canopy untilthe June harvest (Fig. 2). Elevated CO2 increased GPP (+20%)and WUE (+25%) while slightly decreasing ET (−3%) (Fig. 2 andSI Appendix, Table S2). At the June (prestress) harvest, there wasno effect of eCO2 on above-ground biomass. Root growth, how-ever, increased by 77% under elevated CO2 (marginally signifi-cant; SI Appendix, Table S3). Canopy greenness, i.e., the fractionof green vs. total above-ground shoot biomass, increased by 41%(SI Appendix, Table S3 and Fig. S3B), whereas the shoot nitrogencontent (%) and pool (g/m2 ground area) decreased under eCO2relative to aCO2 by 9 and 10%, respectively (Fig. 3B and SIAppendix, Fig. S4 and Table S4).When the ECE was imposed (stress period), the mean daily air

temperature and vapor pressure deficit peaked at 25 °C and 1.8 kPa,respectively (Fig. 1 B and C). These climatic conditions led to agradual reduction in soil moisture (Fig. 1A) and canopy greenness(−10%; SI Appendix, Table S3 and Fig. S3B). NEE and GPP weresignificantly affected by the interaction between eCO2 and ECE, butin a variable manner over time (SI Appendix, Table S2). Comparedwith aCO2 conditions, the eCO2 treatment increased GPP and Reco(by +55 and 23%, respectively) during the first month of drought,but this effect disappeared for GPP during the 2 wk with com-bined drought and heat wave. However, Reco was 35% higher in theeCO2 + ECE treatment relative to aCO2 + ECE (Fig. 2C) resultingin a net release of CO2 from the ecosystem under the eCO2 + ECEtreatment during and immediately after the heat wave (Fig. 2A).During the first month of reduced precipitation, eCO2 buffered thenegative effect of drought on ET and WUE, leading to a higher ET(+6%) and WUE (+51%) under the eCO2 treatment relative toaCO2. However, eCO2 had no lasting impact on ET during andimmediately after the heat wave (Fig. 2 D and E and SI Appendix,Table S2). When precipitation was gradually resumed, in August(Fig. 1A), a progressive recovery of CO2 and water vapor fluxes wasobserved in the macrocosms previously exposed to the ECE. Therecovery of relative soil water moisture was lower in the eCO2 + ECEtreatment relative to aCO2 + ECE (Fig. 1A) due to a strongerrecovery of ET and other fluxes at elevated CO2 (Fig. 2D).During the poststress period (September to November), fluxes

of CO2 and water decreased progressively with decreasing tem-perature and light levels in both CO2 treatments without ECE.The eCO2 treatment significantly increased NEE (+146%), GPP(+46%), Reco (+25%), andWUE (+32%) (Fig. 2 and SI Appendix,Table S2). GPP, NEE, and WUE strongly increased in Septemberand reached values significantly higher than in the treatmentswithout ECE (Fig. 2 and SI Appendix, Table S2). An eCO2 × ECEinteraction affected the NEE and GPP fluxes, suggesting that therecovery of the CO2 fluxes led to significantly higher net and grossecosystem C uptake in the eCO2 + ECE treatment combination(up to 60% and 240% higher GPP and NEE, respectively in theeCO2 + ECE treatment compared with the mean of the remainingtreatment combinations). A time-series analysis of daily CO2fluxes focused on the last two weeks of September shows thatNEE and GPP were significantly affected by the eCO2 × ECE ×Time interaction during this period with P values <0.001 (SIAppendix, Table S5 and Fig. S5), and with the highest C uptake inthe eCO2 + ECE treatment.Above-ground biomass at the November harvest did not differ

between the treatments (Fig. 3A) and confirmed the low pro-ductivity of this grassland type during autumn (20). During thepoststress period, compared with aCO2, eCO2 increased the LeafArea Index (LAI) (+47.9%; SI Appendix, Fig. S3A) and rootgrowth rate (+244%; Fig. 3C) and decreased the shoot nitrogen

Fig. 1. Ecotron-simulated environmental conditions of the year 2050 undertwo atmospheric CO2 concentrations (390 and 520 μmol·mol−1) with andwithout an ECE. (A) Soil moisture relative to field capacity in the top 60 cm.Bars depict rainfall. (B) Experimentally simulated vs. field recorded tempera-ture (daily means) in the heat wave year 2003 and average of the years 1990–2009 (the simulated temperatures at each CO2 concentrations are too similarfor the blue and red lines to be distinct). (C) Simulated air vapor pressuredeficit. (D) Experimentally simulated atmospheric CO2 concentrations. Datarepresent means over 14 (±1) d ± SEM of three replicates. Horizontal barsrepresent the experimental periods (solid green, prestress; solid orange,halved water supply; solid red, no water supply and +3.4 °C temperature in-crease; hatched orange bar, gradual rewetting; hatched green bar, poststress(see Methods for details).

Roy et al. PNAS | May 31, 2016 | vol. 113 | no. 22 | 6225

ECOLO

GY

Dow

nloa

ded

by g

uest

on

May

25,

202

0

content (−21%; SI Appendix, Fig. S4 and Tables S3 and S4). NoCO2 effect was found on canopy greenness or on the shootnitrogen pool. ECE strongly increased the shoot nitrogen pool(+87%, Fig. 3B) relative to the treatment combinations withoutECE (both under aCO2 and eCO2), primarily through an in-crease in the shoot nitrogen content (+53%; SI Appendix, Fig.S4 and Table S4). Belowground, the root growth rate and theroot nitrogen pool were affected by the CO2 × ECE × Timeinteraction (Fig. 3 C and D and SI Appendix, Fig. S6A andTables S3 and S4). We also found an ECE × Time interactionon the root nitrogen content, with a significantly higher contentin the ECE treatment in September (+19%, P1/10 = 0.029; SIAppendix, Fig. S6B).These results suggest that the recovery of CO2 fluxes following

the ECE treatments was stimulated by increased nitrogen avail-ability. This suggestion was confirmed by showing that shoot androot nitrogen pools were predictors of late October CO2 fluxes(GPP and NEE; SI Appendix, Table S6). Furthermore, in agree-ment with this conjecture, we also found that the soil nitrificationpotential measured after the November biomass harvest wastwice as large in the macrocosms that had been exposed toECE compared with macrocosms not exposed to ECE (Fig. 4Aand SI Appendix, Table S7).Integrated over the full growing season (April 26–November 3),

eCO2 significantly increased the cumulative seasonal NEE (+79%),GPP (+30%), and WUE (+32%), whereas the ECE significantlyreduced cumulative NEE (−29%), GPP (−16%), and ET (−18%)(bar plot of Fig. 2 and SI Appendix, Table S7). Interestingly,whereas effects of eCO2 and ECE were not additive, the cumula-tive NEE was 24% higher under the eCO2 + ECE treatmentcompared with aCO2 without ECE (P = 0.027). This result showsthat under the projected climatic conditions of the 2050s, eCO2more than compensated the negative impact of extreme droughtand heat wave. Under aCO2, the ECE reduced the ecosystemC balance (CNEE − Charvested biomass) (Fig. 4B and SI Appendix,Table S7), and turned the functioning of the grassland from a smallC sink (19.9 gC·m−2) to a C source for the atmosphere (−94 gC·m−2),a result consistent with the analysis of the consequence of the sum-mer 2003 heat and drought (5, 21). In contrast, under eCO2, theecosystem C balance increased and reached 115.0 gC·m−2 withoutECE and remaining a C sink, 46.3 gC·m−2, with ECE (Fig. 4B).Most previous studies on the impact of eCO2 associated with

drought and/or elevated temperature did not report whole ecosys-tem responses, but primarily plant responses at various organiza-tional levels. Generally, they show that eCO2 alleviates the stressassociated with drought and/or elevated temperature (4, 16, 18, 22–28), but some studies found inconsistent responses (29–36). Thisvariability could be linked with the seasonality of precipitation (35,37) or with the intensity and length of the imposed stress, becausethe effect of eCO2 has been shown to be positive under moderatedrought and negligible under severe stress when stomata are fullyclosed (14). Owing to high-frequency measurements of ecosystem-level CO2 fluxes, we could detect both mitigation effects of eCO2 onCO2 fluxes during moderate stress and negative effects of eCO2 onNEE via an increased Reco during and immediately after the heatwave. Our results show that these apparently conflicting results,which have been previously documented separately, occur at dif-ferent periods during the growing season and depend indeed on thelevel of water stress. An increase in Reco under eCO2 has beenshown to be prevalent and to result from improved soil watercontent and/or enhanced inputs of labile C that prime the decom-position of soil organic matter (38). We show that the negativeeffect of eCO2 on NEE via increased Reco is transient and had nosubstantial impact on the cumulative net C uptake during thegrowing season. In our experiment, the positive effect of eCO2 onCO2 fluxes under moderate stress cannot be explained through asoil water sparing effect as found in some earlier studies (16, 24),

Fig. 2. Seasonal dynamics (Left, lines) and cumulative fluxes (Right, bars) ofcarbon and water as affected by the CO2 and ECE treatments. (A) NEE. (B) GPP.(C) Reco. (D) ET. (E) WUE. Data represent means over 14 (±1) d ± SEM of threereplicates. Stars indicate significant terms according to repeated-measuresANOVAs [***P < 0.001, ** P < 0.01, * P < 0.05, and (*)P < 0.08]. Different lettersabove bars denote significant differences between individual means. Horizontalbottom bars represent the experimental periods as in Fig. 1.

6226 | www.pnas.org/cgi/doi/10.1073/pnas.1524527113 Roy et al.

Dow

nloa

ded

by g

uest

on

May

25,

202

0

because soil water content was similar with or without the ECE,but rather through the CO2 fertilization effect.

To our knowledge, only a single experiment has so far ad-dressed the consequence of combined eCO2 and elevated tem-perature on the ecosystem C budget (39) and found that thesecombined treatments increased C loss (in one of the four years ofstudy), a result in contradiction to model simulations (40, 41).The most notable effect in that experiment was the increased lossof C under elevated CO2 (through increased heterotrophic soilrespiration) in two of the four years. In our study, we found apositive C budget at the end of the growing season under thecombined eCO2 and ECE treatment, mainly due to a strongrecovery of GPP fluxes in the autumn (post ECE), where theinteraction between eCO2 and ECE led to the highest GPP. Sucha remarkable recovery of GPP under a realistic future scenarioof eCO2 and combined drought and heat wave has not beenreported before, although at species level, it was previously ob-served that eCO2 can enhance nitrogen metabolism and photo-synthesis after drought (42). In our experiment, the enhancedrecovery of GPP under the eCO2 + ECE treatment combinationwas related to increased nitrogen uptake by the newly grownroots and the shoots, which statistically explain the ECE effecton GPP and NEE (SI Appendix, Table S6). Increased plant ni-trogen content has been previously found to be higher undereCO2, elevated temperature, and drought, especially when allthree factors were combined (43). We argue that the unexpectedincrease in canopy and root nitrogen pools (Fig. 3 B and D), anda higher LAI (SI Appendix, Fig. S3A) at the beginning of thepoststress period in the ECE treatments, reflect a higher soilnitrogen availability after the extreme event, as indicated by asignificantly higher microbial nitrification potential in the treat-ments with ECE (Fig. 4A). Recent analyses of 12 decadal ex-periments (44, 45) show that such a positive impact of eCO2 onplant nitrogen acquisition and aboveground net primary pro-duction is not decreasing with time and that the associatedprogressive nitrogen limitation hypothesis (46) is not ubiquitous.In conclusion, our study shows that under realistic extreme events

of heat and drought as predicted for the 2050s, eCO2 does notnecessarily exacerbate the drought and heat stress as sometimessuggested (11, 12), but can contribute to maintain the C sinkfunction of grasslands via an increased root nitrogen uptake whenthe stress is released. Although climatic extremes are likely to be-come more frequent and more severe toward the end of the cen-tury, the current Earth system models do not simulate adequatelythe impacts of climate extremes on biogeochemical cycles (47) andstill have large uncertainties in parameterizing regional responses tosuch events and often provide contrasting results (47–50). Our ex-periment highlights the importance of accounting for interactionsbetween drivers of global change, their seasonality effects, and theimportance of poststress recovery processes for more accuratemodel predictions of the future C budgets of grasslands.

MethodsExtraction and Acclimation of Soil Monoliths. In June 2009, 48 soil monoliths(1 m2 each) including their intact soil and plant communities [dominated byperennial C3 grasses and by the pasture legume Trifolium repens (full speciescomposition; see SI Appendix, Table S1)], were excavated down to the rocklevel (0.6 m depth) from an extensively managed upland seminaturalgrassland site (Redon), near Saint-Genès Champanelle in the French MassifCentral (SI Appendix, Fig. S2). The monoliths were left to recover after theextraction disturbance at the Clermont-Ferrand INRA research station (meanannual temperature 12.4 °C) until September 2009 when they were trans-ported to Montpellier. In April 2010, four randomly chosen 1 m2 monolithswere combined to form one large (4 m2) experimental unit and inserted ineach of the 12 Ecotron macrocosms (SI Appendix, SI Methods). Thereafter,they were left to acclimate to the climatic conditions representative ofyears 2050s and ambient atmospheric CO2 concentrations (390 μmol·mol−1)for 11 mo. The eCO2 treatment (520 μmol·mol−1) started in early March2011 in six randomly chosen macrocosms (the small departure from theset points for 10 d in August was due to a malfunction in the automaticcontrol system). The 30 m3 transparent dome of each macrocosm had a

Fig. 3. Vegetation response variables as affected by the CO2 and ECE treat-ments. Shoot biomass (A) and nitrogen (N) pool in the shoots (B) at the twosuccessive June 9 and November 3 cuts. Total carbon (C) in the new root growth(C) and total pool of N in the new root growth (D) during the prestress andrewetting/poststress periods. Data are means ± SEM of three replicates. Differentletters above bars denote significant differences between individual means.

Roy et al. PNAS | May 31, 2016 | vol. 113 | no. 22 | 6227

ECOLO

GY

Dow

nloa

ded

by g

uest

on

May

25,

202

0

photosynthetically active radiation transmission of 0.86, a figure matchingclosely the annual global radiation ratio between Saint-Genès Champanelleand Montpellier during April to September (0.84).

Simulation of ECE. Climate projections for the original location of thisgrassland for the 2050s (2040–2060) were obtained from the ARPEGEv4atmosphere-ocean general circulation model with the A2-CO2 emissionscenario (51) and using a multivariate statistical downscaling methodology(19) to generate projections over 8 × 8 km grid. The climatic conditions ofan average year of that period were applied from April 2010 to November2, 2011. From June 25 to August 31, 2011, a summer ECE including com-bined drought and heat wave (stress period), simulating the severestevents projected by the downscaled ARPEGEv4 model, was applied tothree randomly selected macrocosms of the six at each CO2 concentration.For the first four weeks of the stress period (June 25–July 21), the irriga-tion amount was reduced by half compared with the control. Then, duringtwo weeks (July 22–August 4), irrigation was stopped and the air tem-perature was increased by 3.4 °C compared with the 2050s average. Thisincrease in air temperature corresponded to 7.1 °C above the 2000–2009average for the same period, a value above the average of the 14 con-secutive hottest days of the exceptional heat wave in summer 2003. FromAugust 5 to 31, irrigation was progressively increased in the treatmentwith ECE to obtain the same cumulative precipitation in the two treatmentsduring that period.

The Gas Exchange System. Each macrocosm unit of the Montpellier Ecotron isan independent open gas exchange measurement system (52), with aconstant flow of air circulating through. The CO2 flux measurement (NEE) isbased on mass balance calculation: The flux of CO2 exchanged between theecosystem and atmosphere is calculated from the difference between thefluxes of CO2 entering and leaving the dome. For this purpose, the CO2 andwater vapor mole fractions are measured at the inlet and outlet (infraredgas analyzers, Licor 7000; Licor), and the mass flow of air at the inletthermal mass flow meter (Sensiflow iG; ABB). Water vapor mole fraction

measurements are used to correct for the dilution of CO2 and the variationof airflow rate at the outlet due to the vapor added or removed in the dome byecosystem transpiration and humidity control. A custom-made manifold system(interconnected solenoid valves, 750 series; Matrix) allows for switching the airfluxes to the infrared gas analyzers from one dome to the other. It takes 12 minto analyze the atmospheric CO2 concentration of the 12 domes.

Missing values or known inaccurate NEE values during unavoidable ex-perimental work (manual watering, mowing, sampling, checks) caused by therespiration of the persons entering the domes were gap-filled by using theequation NEE = f(photosynthetically active radiation), where f is a rectan-gular hyperbola fitted with the data of the day before or after the distur-bance (53). To provide conservative results, all statistics were performedusing only days with less than 33% of gap-filled NEE data. The C flux par-titioning algorithm Reichstein et al. (54) (www.bgc-jena.mpg.de/∼MDIwork/eddyproc/index.php) was used to estimate the daytime Reco.

Other Measurements. Ecosystem evapotranspiration was measured continu-ously by the lysimeter weight changes over time. Four shear beam load cellswith an accuracy of ±200 g (CMI-C3 5,000 kg, Precia-Molen) are used for eachlysimeter. WUE was calculated as mg of CO2 fixed per g of H2O lost. Soilwater content was continuously measured at three soil depths (7, 20, and50 cm) with TDR probes (TRIME Pico 32, IMKO Micromodul-technik, Ger-many Ettlingen). Reported data relate to the 0–60 cm depth and areexpressed relative to soil water content at field capacity. LAI was esti-mated with a sunfleck ceptometer (Decagon Devices) from April to Octo-ber, and canopy greenness was visually estimated every three weeks fromJune to October. Because extensive management was found to be the bestoption for sustaining the production of this type of grassland in the con-text of greater climate variability (20), no fertilization was applied and thenumber of harvests was low, especially in the summer after the start of theECE. The vegetation was cut at 5 cm height on March 14, April 26, June 9,and November 3.

Root growth (0–15 cm depth) was measured monthly using four (8 cmdiameter) ingrowth cores per macrocosm from February to the end of theexperiment. After harvest, shoots and roots were oven-dried (60 °C, 48 h),weighed, and their C and nitrogen content analyzed. Microbial nitrificationpotential was determined on soil samples collected at the end of the ex-periment, according to the method of Lensi et al. (55) described in moredetail by Pinay et al. (56) (SI Appendix, SI Methods).

Statistical Analysis. Linear mixed effects models as available in the R “nlme”package (57) were used to perform repeated-measures ANOVAs on the effectsof CO2, ECE, Time, and their interactions on fortnightly averaged values ofcarbon and water fluxes, with the Ecotron’s macrocosms as a random factoraccounting for temporal pseudoreplication. The statistical models were thensimplified to reach the most parsimonious models by using the automaticmodel simplification “step” procedure based on Akaike’s Information Crite-rion. Planned pairwise comparisons between the means of the treatments atdifferent time periods were performed by using ANOVAs. As we found amarginally significant interaction between CO2 and ECE on NEE and GPPduring the recovery period, additional time-series analyses on daily valueswere performed by using mixed effects models with the Ecotron’s macrocosmsas a random factor. To account for the temporal autocorrelation of the dailyvalues, these models included autoregressive covariance structure (“correla-tion=corAR1(form = ∼1 j macrocosm”) at the macrocosm level. Following theguidelines of Zuur et al. (58), we found that the models with the day-dependent variance coefficients (“varIdent(form =∼ 1 j Day”) fitted bestthe data and were therefore retained in the models (see SI Appendix,Table S4 for the complete R syntax).

ACKNOWLEDGMENTS. Ch. Collin and the experimental field team at Centred’Ecologie Fonctionnelle et Evolutive (CEFE)–CNRS as well as the technicalstaff of INRA Grassland Ecosystem UREP and UERT groups are thanked forextracting the intact soil monoliths. We also thank B. Buatois and the Pla-teforme d’Analyses Chimiques en Ecologie (CEFE and LabEx CEMEB) for thenitrification analyses. This study was supported by the ANR VALIDATE proj-ect grant and the AnimalChange project funded by the European Commu-nity (FP7/2007–2013, Grant 266018). This study benefited from the CNRShuman and technical resources allocated to the ECOTRONS Research Infra-structure as well as from the state allocation “Investissement d’Avenir”AnaEE-France ANR-11-INBS-0001. The Languedoc Roussillon Region andthe Conseil Général de l’Hérault also participated in the funding ofthe Ecotron. A.A. and M.-L.B. received a postdoctoral position throughan INRA scientific package (2010–2014). A.A. was also supported by theEuropean FP7 ExpeER Transnational Access program, and M.B. was addi-tionally supported by the ÖAD WTZ-programme Austria–France and FWFProject P28572-B22.

Fig. 4. Soil nitrification potential at the end of the experiment (A) and fullgrowing season (April 26–November 3) ecosystem carbon (C) balance (>0sink; <0 source) (B) as affected by the CO2 and ECE treatments. Data aremeans ± SEM of three replicates. Different letters above bars denote sig-nificant differences between individual means.

6228 | www.pnas.org/cgi/doi/10.1073/pnas.1524527113 Roy et al.

Dow

nloa

ded

by g

uest

on

May

25,

202

0

1. Jacob D, et al. (2013) EURO-CORDEX: New high-resolution climate change projectionsfor European impact research. Reg Environ Change 14(2):563–578.

2. Christidis N, Jones GS, Stott PA (2015) Dramatically increasing chance of extremely hotsummers since the 2003 European heatwave. Nat Clim Chang 5:46–50.

3. Seneviratne SI, et al. (2012) IPCC 2012: Managing the Risks of Extreme Events andDisasters to Advance Climate Change Adaptation (Cambridge Univ Press, Cambridge,UK), pp 109–230.

4. Reynolds SG, Batello C, Baas S, Mack S (2005) Grassland: A Global Ressource, edGilloway DA (Wageningen Academic Publishers, Wageningen, The Netherlands), pp323–338.

5. Ciais P, et al. (2005) Europe-wide reduction in primary productivity caused by the heatand drought in 2003. Nature 437(7058):529–533.

6. Reichstein M, et al. (2007) Reduction of ecosystem productivity and respiration duringthe European summer 2003 climate anomaly: A joint flux tower, remote sensing andmodelling analysis. Glob Change Biol 13(3):634–651.

7. Reichstein M, et al. (2013) Climate extremes and the carbon cycle. Nature 500(7462):287–295.

8. Frank D, et al. (2015) Effects of climate extremes on the terrestrial carbon cycle:Concepts, processes and potential future impacts. Glob Change Biol 21(8):2861–2880.

9. Zscheischler J, et al. (2014) Extreme events in gross primary production: A charac-terization across continents. Biogeosciences 11:2909–2924.

10. Leakey ADB, et al. (2009) Elevated CO2 effects on plant carbon, nitrogen, and waterrelations: Six important lessons from FACE. J Exp Bot 60(10):2859–2876.

11. Williams AC (2014) Heat and drought extremes likely to stress ecosystem productivityequally or more in a warmer, CO2 rich future. Environ Res Lett 9:101002.

12. Williams IN, Torn MS, Riley WJ, Wehner MF (2014) Impacts of climate extremes ongross primary production under global warming. Environ Res Lett 9:094011.

13. Morgan J, Lecain D, Mosier A, Milchunas D (2001) Elevated CO2 enhances water re-lations and productivity and affects gas exchange in C3 and C4 grasses of the Colo-rado shortgrass steppe. Glob Change Biol 7(4):451–466.

14. Franks PJ, et al. (2013) Sensitivity of plants to changing atmospheric CO2 concentra-tion: From the geological past to the next century. New Phytol 197(4):1077–1094.

15. Holtum JAM, Winter K (2010) Elevated CO2 and forest vegetation: More a water issuethan a carbon issue? Funct Plant Biol 37:694–702.

16. Robredo A, et al. (2007) Elevated CO2 alleviates the impact of drought on barleyimproving water status by lowering stomatal conductance and delaying its effects onphotosynthesis. Environ Exp Bot 59(3):252–263.

17. Albert KR, et al. (2011) Effects of elevated CO₂, warming and drought episodes onplant carbon uptake in a temperate heath ecosystem are controlled by soil waterstatus. Plant Cell Environ 34(7):1207–1222.

18. Morgan JA, et al. (2011) C4 grasses prosper as carbon dioxide eliminates desiccation inwarmed semi-arid grassland. Nature 476(7359):202–205.

19. Boé J, Terray L, Habets F, Martin E (2006) A simple statistical-dynamical downscalingscheme based on weather types and conditional resampling. J Geophys Res 111:D23106.

20. Zwicke M, et al. (2013) Lasting effects of climate disturbance on perennial grasslandabove-ground biomass production under two cutting frequencies. Glob Change Biol19(11):3435–3448.

21. Soussana JF, et al. (2007) Full accounting of the greenhouse gas (CO2, N2O, CH4)budget of nine European grassland sites. Agric Ecosyst Environ 121:121–134.

22. Zinta G, et al. (2014) Physiological, biochemical, and genome-wide transcriptionalanalysis reveals that elevated CO2 mitigates the impact of combined heat wave anddrought stress in Arabidopsis thaliana at multiple organizational levels. Glob ChangeBiol 20(12):3670–3685.

23. Naudts K, Van den Berge J, Janssens IA, Nijs I, Ceulemans R (2013) Combined effects ofwarming and elevated CO2 on the impact of drought in grassland species. Plant Soil369:497–507.

24. Blumenthal DM, et al. (2013) Invasive forb benefits from water savings by nativeplants and carbon fertilization under elevated CO2 and warming. New Phytol 200(4):1156–1165.

25. Miranda-Apodaca J, Pérez-López U, Lacuesta M, Mena-Petite A, Muñoz-Rueda A(2015) The type of competition modulates the ecophysiological response of grasslandspecies to elevated CO2 and drought. Plant Biol (Stuttg) 17(2):298–310.

26. Koutavas A (2013) CO2 fertilization and enhanced drought resistance in Greek firsfrom Cephalonia Island, Greece. Glob Change Biol 19(2):529–539.

27. Dias De Oliveira E, et al. (2013) Can elevated CO2 combined with high temperatureameliorate the effect of terminal drought in wheat? Funct Plant Biol 40(2):160–171.

28. Yu J, Chen L, Xu M, Huang B (2012) Effects of elevated CO2 on physiological responsesof tall fescue to elevated temperature, drought stress, and the combined stresses.Crop Sci 52:1848–1858.

29. Derner JD, et al. (2003) Above- and below-ground responses of C3-C4 species mixturesto elevated CO2 and soil water availability. Glob Change Biol 9(3):452–460.

30. Dukes JS, et al. (2005) Responses of grassland production to single and multiple globalenvironmental changes. PLoS Biol 3(10):e319.

31. Duan H, et al. (2014) Elevated [CO2] does not ameliorate the negative effects of el-evated temperature on drought-induced mortality in Eucalyptus radiata seedlings.Plant Cell Environ 37(7):1598–1613.

32. Godfree RC, et al. (2013) Nonindigenous plant advantage in native and exotic aus-tralian grasses under experimental drought, warming, and atmospheric CO2 enrich-ment. Biology (Basel) 2(2):481–513.

33. Bloor JMG, Pichon P, Falcimagne R, Leadley P, Soussana JF (2010) Effects of warming,summer drought, and CO2 enrichment on aboveground biomass production, flow-ering phenology, and community structure in an upland grassland ecosystem.Ecosystems (N Y) 13(6):888–900.

34. Kongstad J, et al. (2012) High resilience in heathland plants to changes in tempera-ture, drought, and high resilience in heathland plants to changes in temperature,drought, and CO2 in combination: results from the CLIMAITE experiment. Ecosystems(N Y) 15(2):269–283.

35. Brookshire ENJ, Weaver T (2015) Long-term decline in grassland productivity drivenby increasing dryness. Nat Commun 6:7148.

36. Way DA (2011) The bigger they are, the harder they fall: CO2 concentration and treesize affect drought tolerance. Tree Physiol 31(2):115–116.

37. Hovenden MJ, Newton PCD, Wills KE (2014) Seasonal not annual rainfall determinesgrassland biomass response to carbon dioxide. Nature 511(7511):583–586.

38. van Groenigen KJ, Qi X, Osenberg CW, Luo Y, Hungate BA (2014) Faster de-composition under increased atmospheric CO2 limits soil carbon storage. Science344(6183):508–509.

39. Pendall E, et al. (2013) Warming reduces carbon losses from grassland exposed toelevated atmospheric carbon dioxide. PLoS One 8(8):e71921.

40. Luo Y, et al. (2008) Modeled interactive effects of precipitation, temperature, andCO2 on ecosystem carbon and water dynamics in different climatic zones. GlobChange Biol 14(9):1986–1999.

41. Zscheischler J, et al. (2014) Carbon cycle extremes during the 21st century in CMIP5models: Future evolution and attribution to climatic drivers. Geophys Res Lett 41(24):8853–8861.

42. Robredo A, et al. (2011) Elevated CO2 reduces the drought effect on nitrogen me-tabolism in barley plants during drought and subsequent recovery. Environ Exp Bot71(3):399–408.

43. Arndal MF, et al. (2013) Net root growth and nutrient acquisition in response topredicted climate change in two contrasting heathland species. Plant Soil 369(1):615–629.

44. Feng Z, et al. (2015) Constraints to nitrogen acquisition of terrestrial plants underelevated CO2. Glob Change Biol 21(8):3152–3168.

45. Reich PB, Hobbie SE (2012) Decade-long soil nitrogen constraint on the CO2 fertil-ization of plant biomass. Nat Clim Chang 3:278–282.

46. Luo Y, et al. (2004) Progressive nitrogen limitation of ecosystem responses to risingatmospheric carbon dioxide. Bioscience 54(8):731–739.

47. Bahn M, Reichstein M, Dukes JS, Smith MD, McDowell NG (2014) Climate-biosphereinteractions in a more extreme world. New Phytol 202(2):356–359.

48. Smith J, et al. (2007) Projected changes in the organic carbon stocks of croplandmineral soils of European Russia and the Ukraine, 1990-2070. Glob Change Biol 13(2):342–356.

49. Van Oijen M, et al. (2014) Impact of droughts on the carbon cycle in European veg-etation: A probabilistic risk analysis using six vegetation models. Biogeosciences11(22):6357–6375.

50. Soussana JF, Graux AI, Tubiello FN (2010) Improving the use of modelling for pro-jections of climate change impacts on crops and pastures. J Exp Bot 61(8):2217–2228.

51. Naki�cenovi�c N, et al. (2000) IPCC Special Report on Emission Scenarios (CambridgeUniv Press, Cambridge, UK).

52. Milcu A, et al. (2014) Functional diversity of leaf nitrogen concentrations drivesgrassland carbon fluxes. Ecol Lett 17(4):435–444.

53. Casella E, Soussana JF (1997) Long-term effects of CO2 enrichment and temperatureincrease on the carbon balance of a temperate grass sward. J Exp Bot 48(311):1309–1321.

54. Reichstein M, et al. (2005) On the separation of net ecosystem exchange into assim-ilation and ecosystem respiration: Review and improved algorithm. Glob Change Biol11(9):1424–1439.

55. Lensi R, Mazurier S, Gourbiere F, Josserand A (1986) Rapid determination of the ni-trification potential of an acid forest soil and assessment of its variability. Soil BiolBiochem 18(2):239–240.

56. Pinay G, et al. (2007) Impact of atmospheric CO2 and plant life forms on soil microbialactivities. Soil Biol Biochem 39(1):33–42.

57. Pinheiro J, Bates D, DebRoy S, Sarkar D, R Development Core Team (2016) nlme:Linear and Nonlinear Mixed Effects Models. R Package version 3.1–97. Available atCRAN.R-project.org/package=nlme. Accessed April 2016.

58. Zuur AF, Ieno EN, Walker N, Saveliev AA, Smith GM (2009) Mixed Effects Models andExtensions in Ecology with R (Springer, New York), 10.1007/978-0-387-87458-6.

Roy et al. PNAS | May 31, 2016 | vol. 113 | no. 22 | 6229

ECOLO

GY

Dow

nloa

ded

by g

uest

on

May

25,

202

0